Internal documents reveal OpenAI employs hundreds of contractors to evaluate anonymized ChatGPT conversations. This initiative, codenamed Project Lily, tasks human reviewers with assessing user prompts and model responses.
Reviewers aim to enhance model helpfulness and reduce undesirable behaviors like anthropomorphism. Although OpenAI filters for personally identifiable information, reviewed logs may still contain sensitive user details.
OpenAI enables data sharing for model improvement by default for consumer plans, including Plus and Pro tiers. Users must manually opt-out through data control settings to prevent future conversations from being reviewed.